A joint extreme value and mixed copula framework for modeling blood pressure extremes and BMI association

Overweight and hypertension are two critical cardiovascular risk factors, making the relationship between body mass index (BMI) and blood pressure (BP) particularly important for early prevention. Existing approaches that model systolic blood pressure (SBP) and diastolic blood pressure (DBP) separately, or reduce BP to a binary hypertension indicator, either ignore their inherent physiological correlation or discard clinically relevant continuous information. Furthermore, reliance on mean-based regression fails to capture the extreme elevations most critical for cardiovascular risk assessment. We propose a framework combining extreme value theory and mixed copula modeling to construct a joint, continuous BP index. Marginal distributions of SBP and DBP were fitted using a Lognormal-Generalized Pareto distribution composite model, and their dependence structure was captured via mixed copulas. The resulting BP index was regressed on BMI using weighted least squares. The framework was applied to 41,451 health examination records from Suzhou, China (2021-2023), stratified by age and sex, with 39,740 from 2011-2013 and 114,096 records from 2016–2018 serving as a temporal reference. The Lognormal–Generalized Pareto distribution composite model provided a superior fit to the empirical SBP and DBP distributions compared with single parametric distributions across all age and sex subgroups. Mixed copulas outperformed single copulas in capturing dependence structure in all subgroups. In the most recent 2021–2023 cohort, younger and middle-aged adults exhibited strong upper-tail dependence (indicating substantial co-movement of extreme SBP and DBP), whereas older adults demonstrated a progressive decoupling of the two measurements. Within this same period, BMI was positively associated with the BP index in nearly all subgroups. Across the three study periods, both the dependence structures and BMI coefficients showed heterogeneous and non-monotonic patterns, with significant reductions in the BMI coefficients in 2021–2023 concentrated primarily among several male age groups. The proposed framework captures marginal tail behavior and joint SBP–DBP dependence and provides a continuous summary of concurrent BP elevation. The age–sex-specific results revealed heterogeneous dependence structures and BMI associations across demographic groups and study periods. These findings highlight the potential value of incorporating tail and dependence modeling into epidemiological analyses, and future research should validate the framework in independent prospective cohorts.